Infrastructure Error Real
Warfare, Governance, and the Gap Between AI's Promise and Its Physics
The cyberattack on civilian alert apps in Ukraine and Russia fits a documented pattern: ransomware attacks on water treatment facilities, power grid intrusions attributed to state actors across multiple countries, and now the targeting of civilian emergency notification systems. Each represents a category of harm distinct from traditional kinetic warfare, and the laws of armed conflict — largely written in the mid-twentieth century — have not kept pace with cyberattacks on civilian warning infrastructure as a legally or strategically defined category.
Musk's SpaceX orbital compute ambitions introduce a related governance gap at a different altitude — literally. The International Telecommunication Union governs radio frequency allocation and orbital slots, but its mandate does not extend to the computing payloads satellites might carry. If SpaceX deploys significant orbital compute capacity that is available for commercial use by any paying global customer, a nation-state could theoretically use that capacity for signals intelligence processing or military logistics planning entirely outside existing treaty frameworks. Starlink's role in the Ukraine-Russia conflict — providing communications infrastructure that proved difficult to jam — is already the first major case study in what commercial space infrastructure means for modern conflict, and the strategic lessons from it have not been fully drawn.
The Grok cost problem and the GPT-6 Astra 'vibe manufacturing' phenomenon, read together, describe where the AI industry actually sits versus where its marketing claims it is. The ceiling of capability is genuinely higher than it was two years ago. The floor — reliability, error rate in consequential applications, cost per agent run — has improved but remains a source of real friction for users who encounter it through failed physical prototypes or unexpected usage caps. Musk's public acknowledgment that Grok's agent economics are unsustainable is either unusual candor or a strategic move to pre-empt bad news — and plausibly both.
The 'vibe manufacturing' workflow is meaningful precisely because it is approximate rather than precise. Hobbyists are describing a process of describing intent, iterating conversationally, and arriving at a workable result without formal CAD training. The historical barrier to physical making — learning complex software, mastering precise specification requirements — has been meaningfully lowered. The path to serious industrial application runs directly through dramatically better error detection before the manufacturing stage. Until that reliability threshold is reached, the workflow scales for hobbyist prototyping and stalls at the door of anything where tolerances carry real stakes.